Method and Apparatus for Object Re-identification

1. A method for object re-identification, comprising, for each pair of objects: selecting one image from each of two sets of images associated with two objects in the pair of objects respectively, to constitute a pair of current images for the pair of objects; inputting the pair of current images to a preconfigured feature extraction network, to obtain feature information for the pair of current images; determining whether the pair of objects are one and the same object based on the feature information for the pair of current images and feature information for one or more pairs of historical images for the pair of objects by using a preconfigured re-identification network, comprising: calculating input information to the re-identification network based on the feature information for the pair of current images and the feature information for the one or more pairs of historical images for the pair of objects; and inputting the input information to the re-identification network to obtain a determination result as to whether the pair of objects are one and the same object; and outputting the determination result of said determining when the determination result is that the pair of objects are one and the same object or that the pair of objects are not one and the same object, or repeating the above steps using the pair of current images as a pair of historical images for the pair of objects when the determination result is that it is uncertain whether the pair of objects are one and the same object, wherein said calculating the input information to the re-identification network based on the feature information for the pair of current images and the feature information for the one or more pairs of historical images for the pair of objects comprises one of: dividing the images in the pair of current images and the one or more pairs of historical images for the pair of objects into two sets of images in accordance with their associated objects, and pairing every two images in the two sets of images to obtain a plurality of regrouped pairs of images and calculating a matching value between respective features of two images in each of the plurality of regrouped pairs of images, determining a distance feature based on the matching values for all of the plurality of regrouped pairs of images, and determining the distance feature as the input information; or calculating differences between respective features of two images in the pair of current images to obtain a feature difference vector for the pair of current images, dividing the images in the pair of current images and the one or more pairs of historical images for the pair of objects into two sets of images in accordance with their associated objects, pairing every two images in the two sets of images to obtain a plurality of regrouped pairs of images, calculating a matching value between respective features of two images in each of the plurality of regrouped pairs of images, determining a distance feature based on the matching values for all of the plurality of regrouped pairs of images and determining the feature difference vector and the distance feature as the input information; or calculating a weighted average of feature difference vectors for the one or more pairs of historical images for the pair of objects to obtain a historical feature difference vector, dividing the images in the pair of current images and the one or more pairs of historical images for the pair of objects into two sets of images in accordance with their associated objects, pairing every two images in the two sets of images to obtain a plurality of regrouped pairs of images, calculating a matching value between respective features of two images in each of the plurality of regrouped pairs of images, determining a distance feature based on the matching values for all of the plurality of regrouped pairs of images, and determining the historical feature difference vector and the distance feature as the input information; or calculating differences between respective features of two images in the pair of current images to obtain a feature difference vector for the pair of current images, calculating a weighted average of feature difference vectors for the one or more pairs of historical images for the pair of objects to obtain a historical feature difference vector, dividing the images in the pair of current images and the one or more pairs of historical images for the pair of objects into two sets of images in accordance with their associated objects, pairing every two images in the two sets of images to obtain a plurality of regrouped pairs of images, calculating a matching value between respective features of two images in each of the plurality of regrouped pairs of images, determining a distance feature based on the matching values for all of the plurality of regrouped pairs of images, and determining the feature difference vector, the historical feature difference vector and the distance feature as the input information.

2. The method of claim 1 , wherein the matching value is a distance value and said calculating the matching value between the respective features of the two images in each of the plurality of regrouped pairs of images and determining the distance feature based on the matching values for all of the plurality of regrouped pairs of images comprise: calculating a distance value between the respective features of the two images in each of the plurality of regrouped pairs of images; selecting a maximum distance value and a minimum distance value among the distance values for all the plurality of regrouped pairs of images, and calculating an average of the distance values for all the plurality of regrouped pairs of images to obtain an average distance value; and determining the maximum distance value, the minimum distance value and the average distance value as the distance feature.

3. The method of claim 1 , wherein the matching value is a dot product value and said calculating the matching value between the respective features of the two images in each of the plurality of regrouped pairs of images and determining the distance feature based on the matching values for all of the plurality of regrouped pairs of images comprise: calculating a dot product value between the respective features of the two images in each of the plurality of regrouped pairs of images; selecting a maximum dot product value and a minimum dot product value among the dot product values for all the plurality of regrouped pairs of images, and calculating an average of the dot product values for all the plurality of regrouped pairs of images to obtain an average dot product value; and determining the maximum dot product value, the minimum dot product value and the average dot product value as the distance feature.

4. The method of claim 1 , wherein the matching value comprises a distance value and a dot product value and said calculating the matching value between the respective features of the two images in each of the plurality of regrouped pairs of images and determining the distance feature based on the matching values for all of the plurality of regrouped pairs of images comprise: calculating a distance value between the respective features of the two images in each of the plurality of regrouped pairs of images, selecting a maximum distance value and a minimum distance value among the distance values for all the plurality of regrouped pairs of images, and calculating an average of the distance values for all the plurality of regrouped pairs of images to obtain an average distance value; calculating a dot product value between the respective features of the two images in each of the plurality of regrouped pairs of images, selecting a maximum dot product value and a minimum dot product value among the dot product values for all the plurality of regrouped pairs of images, and calculating an average of the dot product values for all the plurality of regrouped pairs of images to obtain an average dot product value; and determining the maximum distance value, the minimum distance value, the average distance value, the maximum dot product value, the minimum dot product value and the average dot product value as the distance feature.

5. The method of claim 1 , wherein said inputting the input information to the re-identification network to obtain the determination result as to whether the pair of objects are one and the same object comprises: inputting the input information to the re-identification network; and calculating for the pair of objects, by using the re-identification network, an evaluation value for each of a result that the pair of objects are one and the same object, a result that the pair of objects are not one and the same object or a result that it is uncertain whether the pair of objects are one and the same object, and determining one of the results that has the largest evaluation value as the determination result.

6. The method of claim 1 , wherein the re-identification network has a structure comprising sequentially an input layer, two fully connected layers and an output layer.

7. An apparatus for object re-identification, comprising a selecting unit, a feature extracting unit, a determining unit and a processing unit, wherein, for each pair of objects: the selecting unit is configured to select one image from each of two sets of images associated with two objects in the pair of objects respectively, to constitute a pair of current images for the pair of objects; the feature extracting unit is configured to input the pair of current images to a preconfigured feature extraction network, to obtain feature information for the pair of current images; the determining unit is configured to determine whether the pair of objects are one and the same object based on the feature information for the pair of current images and feature information for one or more pairs of historical images for the pair of objects by using a preconfigured re-identification network; and the processing unit is configured to output a determination result obtained by the determining unit when the determination result is that the pair of objects are one and the same object or that the pair of objects are not one and the same object, or use the pair of current images as a pair of historical images for the pair of objects and trigger the selecting unit when the determination result is that it is uncertain whether the pair of objects are one and the same object, wherein the determining unit comprises a calculating sub-unit configured to calculate input information to the re-identification network based on the feature information for the pair of current images and the feature information for the one or more pairs of historical images for the pair of objects, and a determining sub-unit configured to input the input information to the re-identification network to obtain the determination result as to whether the pair of objects are one and the same object, and wherein the calculating sub-unit is configured to perform at least one of: divide the images in the pair of current images and the one or more pairs of historical images for the pair of objects into two sets of images in accordance with their associated objects, pair every two images in the two sets of images to obtain a plurality of regrouped pairs of images, calculate a matching value between respective features of two images in each of the plurality of regrouped pairs of images, determine a distance feature based on the matching values for all of the plurality of regrouped pairs of images, and determine the distance feature as the input information; or calculate differences between respective features of two images in the pair of current images to obtain a feature difference vector for the pair of current images, divide the images in the pair of current images and the one or more pairs of historical images for the pair of objects into two sets of images in accordance with their associated objects, pair every two images in the two sets of images to obtain a plurality of regrouped pairs of images, calculate a matching value between respective features of two images in each of the plurality of regrouped pairs of images, determine a distance feature based on the matching values for all of the plurality of regrouped pairs of images, and determine the feature difference vector and the distance feature as the input information; or calculate a weighted average of feature difference vectors for the one or more pairs of historical images for the pair of objects to obtain a historical feature difference vector, divide the images in the pair of current images and the one or more pairs of historical images for the pair of objects into two sets of images in accordance with their associated objects, pair every two images in the two sets of images to obtain a plurality of regrouped pairs of images, calculate a matching value between respective features of two images in each of the plurality of regrouped pairs of images, determine a distance feature based on the matching values for all of the plurality of regrouped pairs of images, and determine the historical feature difference vector and the distance feature as the input information; or calculate differences between respective features of two images in the pair of current images to obtain a feature difference vector for the pair of current images, calculate a weighted average of feature difference vectors for the one or more pairs of historical images for the pair of objects to obtain a historical feature difference vector, divide the images in the pair of current images and the one or more pairs of historical images for the pair of objects into two sets of images in accordance with their associated objects, pair every two images in the two sets of images to obtain a plurality of regrouped pairs of images, calculate a matching value between respective features of two images in each of the plurality of regrouped pairs of images, determine a distance feature based on the matching values for all of the plurality of regrouped pairs of images, and determine the feature difference vector, the historical feature difference vector and the distance feature as the input information.

8. The apparatus of claim 7 , wherein the matching value is a distance value and the calculating sub-unit is configured to calculate the matching value between the respective features of the two images in each of the plurality of regrouped pairs of images and determine the distance feature based on the matching values for all of the plurality of regrouped pairs of images by: calculating a distance value between the respective features of the two images in each of the plurality of regrouped pairs of images; selecting a maximum distance value and a minimum distance value among the distance values for all the plurality of regrouped pairs of images, and calculating an average of the distance values for all the plurality of regrouped pairs of images to obtain an average distance value; and determining the maximum distance value, the minimum distance value and the average distance value as the distance feature.

9. The apparatus of claim 7 , wherein the matching value is a dot product value and the calculating sub-unit is configured to calculate the matching value between the respective features of the two images in each of the plurality of regrouped pairs of images and determine the distance feature based on the matching values for all of the plurality of regrouped pairs of images by: calculating a dot product value between the respective features of the two images in each of the plurality of regrouped pairs of images; selecting a maximum dot product value and a minimum dot product value among the dot product values for all the plurality of regrouped pairs of images, and calculating an average of the dot product values for all the plurality of regrouped pairs of images to obtain an average dot product value; and determining the maximum dot product value, the minimum dot product value and the average dot product value as the distance feature.

10. The apparatus of claim 7 , wherein the matching value comprises a distance value and a dot product value and the calculating sub-unit is configured to calculate the matching value between the respective features of the two images in each of the plurality of regrouped pairs of images and determine the distance feature based on the matching values for all of the plurality of regrouped pairs of images by: calculating a distance value between the respective features of the two images in each of the plurality of regrouped pairs of images, selecting a maximum distance value and a minimum distance value among the distance values for all the plurality of regrouped pairs of images, and calculating an average of the distance values for all the plurality of regrouped pairs of images to obtain an average distance value; calculating a dot product value between the respective features of the two images in each of the plurality of regrouped pairs of images, selecting a maximum dot product value and a minimum dot product value among the dot product values for all the plurality of regrouped pairs of images, and calculating an average of the dot product values for all the plurality of regrouped pairs of images to obtain an average dot product value; and determining the maximum distance value, the minimum distance value, the average distance value, the maximum dot product value, the minimum dot product value and the average dot product value as the distance feature.

11. The apparatus of claim 7 , wherein the determining sub-unit is configured to: input the input information to the re-identification network; and calculate for the pair of objects, by using the re-identification network, an evaluation value for each of a result that the pair of objects are one and the same object, a result that the pair of objects are not one and the same object or a result that it is uncertain whether the pair of objects are one and the same object, and determine one of the results that has the largest evaluation value as the determination result.

12. The apparatus of claim 7 , wherein the re-identification network has a structure comprising sequentially an input layer, two fully connected layers and an output layer.

13. The method of claim 1 , wherein the one image is selected randomly or according to a chronological order.

14. The method of claim 2 , wherein the distance value is one of: a Euclidean distance, a Manhattan distance, a Chebyshev distance, a Minkowski distance, a cosine distance, or a Mahalanobis distance.

15. The method of claim 1 , wherein the feature extraction network is obtained by training a model based on a neural network that is capable of extracting features from images.

16. The method of claim 1 , comprising determining whether a predetermined end condition is satisfied when the determination result is that it is uncertain whether the pair of objects are one and the same object.

17. The apparatus of claim 7 , wherein the one image is selected randomly or according to a chronological order.

18. The apparatus of claim 8 , wherein the distance value is one of: a Euclidean distance, a Manhattan distance, a Chebyshev distance, a Minkowski distance, a cosine distance, or a Mahalanobis distance.

19. The apparatus of claim 7 , wherein the feature extraction network is obtained by training a model based on a neural network that is capable of extracting features from images.

20. The apparatus of claim 7 , wherein the determining sub-unit is configured to determine whether a predetermined end condition is satisfied when the determination result is that it is uncertain whether the pair of objects are one and the same object.